University of Glasgow
Stratifying risk for covert atrial fibrillation after stroke: a multimodal analysis of clinical, biomarker, and AI tools
Abstract
dc:description.abstractIn short, I explored strategies to improve AF detection following ischaemic stroke or TIA, focusing on identifying patients unlikely to benefit from prolonged CRM. A systematic review confirmed a dose–response relationship between monitoring duration and AF detection, with the highest yields observed after 30 days or more of continuous monitoring. Through a series of retrospective and prospective cohort studies, I demonstrated that the integration of clinical variables, ECG parameters, natriuretic peptides, and AI–based models enhances risk prediction for the absence of AFDAS. I showed that younger age, absence of lipid-lowering therapy, shorter QT interval, and higher Q offset on ECG are independent predictors of AFDAS absence, forming the basis of the PRECISE risk score. I further demonstrated that the addition of MR-proANP and NT-proBNP levels improves the discrimination of clinical risk models, with MR-proANP achieving an AUROC of 0.80 and reducing unnecessary monitoring by more than 25%. In a separate analysis of 12-lead ECGs from 3,966 stroke patients in SR, I evaluated CNNs and AutoML frameworks for AF prediction through collaborations with the University of Liverpool. ECG-based models alone had limited discriminatory performance (AUROC, 0.51), but models incorporating clinical data achieved AUROC values up to 0.70, highlighting the benefit of multimodal approaches. These findings support a more nuanced and resource-efficient approach to poststroke AF screening. They advocate for the targeted use of prolonged CRM in highrisk individuals and reinforce the need for external validation of integrated biomarker and AI-based tools. Future work should focus on refining prediction models in diverse populations, exploring wearable technologies, and integrating AI into routine care to guide personalised secondary prevention.
Degree
thesis:*- Level dc:type.qualificationlevel
- PhD
- Grantor dc:publisher.institution
- University of Glasgow
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Katsas, Georgios
Subjects
dc:subject × 2Rights
- Language dc:language
- en
Identifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.5525/gla.thesis.86085